The specific problem I'm thinking of is this:
- weather has to be synchronised across clients (if you see it's foggy and I don't, that's very bad)
- weather has to exist on billions of procgen planets
- weather therefore has to be consistent and persistent to stop weird stuff happening when two people get close enough together to merge instances (and also stop "I don't like fog" / relog / "ah, bright sunshine")
- it also has, as you say, to be consistent between the orbital level flying right down (in normal space, potentially) to the surface through the atmosphere layers
...all things which they've solved for terrain.
- weather
changes over time
...so the generation has to be dynamic and very efficient ... and ideally, given the rest of the game, give reasonably realistic weather patterns for the terrain, wind going around obstacles, etc.
The combination of the three underlined bits is the bit I cannot possibly see how they can do. Current stellar forge surface generation takes a few seconds just for a heightmap. I can just about see the very edge of how they might theoretically get something that works as a proof of concept for gas-giant atmospheres (where the lack of terrain simplifies matters a lot). Now, of course, Frontier can hire someone much better than me at "procedural weather generation" to work on the problem. I hope they do, because it would be amazing. But if the next five years is simply and only "Elite: Weather" that might not be enough.
The other games are all single-player (or limited location multiplayer?), so "consistent and persistent" isn't needed.